From aab3bc7d9e7bc8e061677f4d96ea1c6e444c9a87 Mon Sep 17 00:00:00 2001
From: jaxperro <jaxtorres@gmail.com>
Date: Thu, 23 Jul 2026 15:10:38 -0400
Subject: [PATCH] =?UTF-8?q?research:=20miss=5Faudit=20(T8)=20=E2=80=94=20c?=
 =?UTF-8?q?hain-only=20guard=20counterfactual=20(v0=20fallback=20retracted?=
 =?UTF-8?q?=20in-file)?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

98 chain-graded misses. The useful inversion: crater_reject misses were
GOOD misses (-$7.74/-$6.32 per $100 both books) — unfillable entries were
adversely selected, consistent with the week's winner's-curse theme, and
puts a question mark on the 07-20 requote-retry's EV (follow-up: score
retry outcomes). Everything else outlier-dominated at n<=21/bucket (one
+1567% longshot carries depth_thin); no filter changes warranted. Ledger
deepens daily — maker-window expiries now feed it.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
---
 research/miss_audit.py | 112 +++++++++++++++++++++++++++++++++++++++++
 1 file changed, 112 insertions(+)
 create mode 100644 research/miss_audit.py

diff --git a/research/miss_audit.py b/research/miss_audit.py
new file mode 100644
index 00000000..24e642b3
--- /dev/null
+++ b/research/miss_audit.py
@@ -0,0 +1,112 @@
+#!/usr/bin/env python3
+"""T8 EXPLORATORY (2026-07-23) — guard counterfactual audit: every recorded
+miss, scored to chain truth, bucketed by the guard that caused it. Which
+gates discard positive EV and which earn their keep?
+
+Sources: copybot state missed lists (both books) + missed_archive spools
+where present. Chain-true where token resolvable (payouts_for); falls back
+to the bot's own settled would-be pnl otherwise. $100-normalized EV per
+guard bucket + realized would-be at recorded stakes."""
+import json
+import os
+import re
+import sys
+
+sys.path.insert(0, os.path.dirname(os.path.abspath(__file__)))
+import tape                                    # noqa: E402
+import forward as fwd                          # noqa: E402
+
+ROOT = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
+BUCKETS = [
+    ("event_cap", r"event cap"),
+    ("price_guard", r"price moved"),
+    ("depth_thin", r"thin book|depth cap|dust"),
+    ("no_ask", r"no ask"),
+    ("crater_reject", r"order rejected|no orders found"),
+    ("hold_expired", r"hold expired|expired unfilled"),
+    ("maker_unfilled", r"maker bid unfilled"),
+    ("backfill_seed", r"held before we started"),
+]
+
+
+def bucket(reason):
+    r = (reason or "").lower()
+    for tag, pat in BUCKETS:
+        if re.search(pat, r):
+            return tag
+    return "other"
+
+
+def load():
+    rows = []
+    for st_path, arch, book in (
+            ("copybot_state.json", "copybot_missed_archive.jsonl", "paper"),
+            ("copybot_state.live.json", "copybot_missed_archive.live.jsonl",
+             "live")):
+        try:
+            for m in json.load(open(os.path.join(ROOT, st_path))).get(
+                    "missed", []):
+                rows.append({**m, "_book": book})
+        except FileNotFoundError:
+            pass
+        ap = os.path.join(ROOT, arch)
+        if os.path.exists(ap):
+            for ln in open(ap):
+                try:
+                    rows.append({**json.loads(ln), "_book": book})
+                except Exception:
+                    pass
+    return rows
+
+
+def main():
+    rows = load()
+    print(f"missed rows: {len(rows)} "
+          f"(paper {sum(1 for r in rows if r['_book']=='paper')} / "
+          f"live {sum(1 for r in rows if r['_book']=='live')})")
+    db = tape.connect()
+    tape.build_resolved(db)
+    toks = [str(r.get("token")) for r in rows if r.get("token")]
+    pays = fwd.payouts_for(db, toks)
+    graded = []
+    for r in rows:
+        # CHAIN-ONLY (v0 fallback reconstructed payout from mark-based
+        # miss-pnl and inflated every riser into a 'win' — retracted)
+        pay = pays.get(str(r.get("token"))) if r.get("token") else None
+        if pay is None or pay == 0.5 or not r.get("price"):
+            continue
+        graded.append((r, pay))
+    print(f"chain/bot-graded: {len(graded)}")
+
+    def line(tag, rs):
+        if not rs:
+            return
+        n = len(rs)
+        ev = sum(100.0 / r["price"] * (p - r["price"]) for r, p in rs) / n
+        hit = sum(1 for _, p in rs if p == 1.0) / n
+        stake_pnl = sum((r.get("stake", 0) / r["price"]) * (p - r["price"])
+                        for r, p in rs)
+        px = sum(r["price"] for r, _ in rs) / n
+        top = sorted((100.0 / r["price"] * (p - r["price"]) for r, p in rs),
+                     reverse=True)[:3]
+        print(f"  {tag:<15} n={n:<5} EV/$100 {ev:+7.2f} · hit {hit:.2f} · "
+              f"avg px {px:.2f} · top3 {['%+.0f' % t for t in top]} · "
+              f"at real stakes {stake_pnl:+9.2f}")
+
+    for book in ("paper", "live"):
+        rs = [(r, p) for r, p in graded if r["_book"] == book]
+        print(f"\n== {book.upper()} ==")
+        line("ALL MISSES", rs)
+        tags = sorted({bucket(r.get("reason")) for r, _ in rs})
+        for t in tags:
+            line(t, [(r, p) for r, p in rs if bucket(r.get("reason")) == t])
+    from collections import Counter
+    census = Counter((r.get("reason") or "")[:44] for r, _ in graded
+                     if bucket(r.get("reason")) == "other")
+    print("\n'other' reason census:")
+    for why, n in census.most_common(6):
+        print(f"  {n:>4}  {why}")
+
+
+if __name__ == "__main__":
+    main()